Hi!

I don't know if this behaviour is intentional but I think it is faulty behaviour. If we disable wsrep_on on a Galera node, restart it, perform changes and then rejoin it to the cluster, the changes are not overwritten or replicated.

Steps to reproduce:
1. Create a galera cluser with 2 or moren odes
2. Shutdown 1 node
3. Change that node's configuration to wsrep_on=0
4. Start the node, insert a row into a table
5. Stop the node, wsrep_on = 1, start the node
6. Delete the changes inserted in step 4
7. Observer the other nodes go down

In my opinion, the server should remove / invalidate grastate.dat when it is started without galera being enabled. Otherwise, if the joining node has different data, this could result in very bad results:

• The other nodes would go down when this node receives changes on the incorrect data
• This node would go down (and still overwrite these faulty changes) if the other nodes cause for example a primary or unique key conflict, causing an unexpected outage

Thanks!
Michaël